InsuranceGoal: find where AI pays off2-week review + 9 weeks
A stalled chatbot pilot, turned into a tool brokers use daily
A chatbot pilot with another vendor had been “almost ready” for six months. Brokers did not use it. The broker asked us whether to kill it or save it.
- Client
- Insurance broker for SMEs and fleets, Poland
- Their team
- COO (owner), 4 senior brokers, IT manager
- Our team
- Engagement lead, 2 engineers
- Timeline
- 2-week review · 9 weeks to live
2.5 h → 35 min
to prepare a comparison offer
40 brokers
use it every day
12 insurers
compared in one view
+60%
offers sent per broker per week
The business goal
Let brokers send more offers in less time — especially for fleet and SME property insurance, where comparing insurers’ terms eats hours.
Where it was
For every offer, brokers opened terms and tariffs from up to 12 insurers — PDFs and online portals — and copied the important conditions into Excel by hand. The chatbot answered general questions nobody was asking.
12 insurers
PDF terms · tariffs · portals
Broker
reads and copies into Excel
Offer
2.5 h later
What we did
- 1Review of what exists · 2 weeksThe expensive part of a broker’s day was comparison, not questions. We kept the old pilot’s document ingestion and dropped the chat.
- 2Outcome brief · 1 weekTarget and exit rule agreed with the COO and four senior brokers.
- 3Pilot · 6 weeksAn assistant that pulls the selected insurers’ terms into a comparison table, each line linked to its source clause. 13 brokers used it voluntarily by week 5 — continue.
- 4Production · 3 weeksPlug-in for the CRM, offer document template, automatic updates when insurers change their terms.
What did not work
Automatic premium calculation: insurers’ online calculators change too often and two of them block automated access. Brokers still enter premiums themselves — the tool does everything else.
Who did what
The client’s team
Senior brokers defined what a good comparison looks like; IT secured access to insurers’ documents.
Our team
Re-scoped the pilot, built the assistant, measured time per offer.
Together
The decision to stop the chatbot — the hardest and most useful one.
Before and after
Under the hood
Ingestion of insurers’ policy wordings and tariffs (reused from the old pilot); retrieval with citations to the exact clause; structured comparison table; CRM plug-in; offer document generation; weekly re-ingestion with change alerts.
The most valuable thing they did in the first two weeks was tell us to stop building a chatbot.